A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 14, 05:30 UTC
[CI] Pooling models mteb test disable enforce_eager (#26408)
Signed-off-by: wang.yuqi <noooop@126.com>
wang.yuqi committed
Oct 8, 2025 at 20:15 UTC
e39dc46f8fe61803032a5f51ba76f8fa03ba0b41
2 files changed
-3
tests/models/language/generation_ppl_test/ppl_utils.py
-1
@@ -51,7 +51,6 @@ def wikitext_ppl_test(
51
gpu_memory_utilization=0.7,
52
max_model_len=max_length,
53
max_num_seqs=1,
54
- enforce_eager=True,
54
**vllm_extra_kwargs,
55
) as vllm_model:
56
# Use max_num_seqs=1 to avoid OOM,
tests/models/language/pooling_mteb_test/mteb_utils.py
-2
@@ -192,7 +192,6 @@ def mteb_test_embed_models(
192
model_info.name,
193
runner="pooling",
194
max_model_len=None,
195
- enforce_eager=True,
195
**vllm_extra_kwargs,
196
) as vllm_model:
197
model_config = vllm_model.llm.llm_engine.model_config
@@ -349,7 +348,6 @@ def mteb_test_rerank_models(
348
runner="pooling",
349
max_model_len=None,
350
max_num_seqs=8,
352
- enforce_eager=True,
351
**vllm_extra_kwargs,
352
) as vllm_model:
353
model_config = vllm_model.llm.llm_engine.model_config